Founding Research Engineer

Mederva is an AI longevity lab on a mission to add healthy years to everyone's life. We help people halt, and even reverse, the progression of chronic disease and slow age-related decline, using AI and proven behavioral science. The proactive care that the wealthy get from $100K concierge clinics shouldn't be reserved for the few. So we partner with healthcare providers to offer our AI and clinician platform to their patients, covered by insurance. We're physician-founded, with a team from Meta Superintelligence Labs, Headspace, Levels Health and Stanford. Medicare's Innovation Center selected us for its new ACCESS program, which pays for health outcomes rather than services. That makes patient outcomes our business's only objective function. We're backed by YC, HF0, Eniac Ventures and Wing VC, and revenue has grown 10x in four months. We're hiring our first AI researcher to found the lab. Most RL environments are simulated. Ours is real, and on track to millions of intervention and outcome pairs within the next year, each grounded in a patient's full medical history and tracked over months and years. You'll design and build that environment and the agents that learn from it. What you'll own This is our first AI hire. You'll start the lab, set its technical direction, and shape the team that grows around it. You'll own the agents, evals and learning loops that turn longitudinal patient data into action, and you'll publish benchmarks and models. The core problem: given everything we know about a patient, what should happen next, what can an agent safely do about it, and did it actually work? The loop we care about: patient state → agent decision → intervention → outcome → better agent. Clinicians stay in the loop, and every improvement ships to real patients. You'll build systems that:

• Turn fragmented clinical, behavioral and conversational history into an evolving model of each patient.
• Give agents the context, memory, tools, verification and escalation paths to act reliably across days and months.
• Build evals from real failures, then use them to improve context, tools, orchestration, model choice and post-training.
• Connect agent decisions to what happened next: patient behavior, clinical follow-through, and outcomes.
• Run at production scale with low latency and cost.

Who we're looking for

• You've shipped frontier AI systems with real-world impact. You can explain how they failed, how you measured it, what you changed, and whether it helped.
• Breadth across the agent stack: architecture, context engineering, memory, retrieval, tool use, model routing, evals, guardrails and observability.
• Strong engineering fundamentals. You own systems end to end, move without a spec, and prefer simple experiments over elaborate architecture.
• Strong plus: post-training (fine-tuning, RL, reward modeling, outcome-based training) and early-stage or founding experience.
• Healthcare experience is not required.

Compensation $250K–$285K base + 0.5%–1.0% equity. San Francisco Bay Area preferred; remote US considered. Unfortunately, scammers sometimes pose as recruiters from companies like ours. All legitimate communication from our team will come from a @medervahealth.com email address.